Do 90% of ChatGPT Citations Come From Pages Ranked 21 or Lower on Google?

Where does the 90% figure come from?

The figure comes from Semrush's study of how AI search could affect traffic, published on 21 July 2025. It looked at AI Overviews, AI Mode, ChatGPT, Claude and Perplexity across more than 500 high-value digital marketing and SEO topics, turned into search terms and prompts. Its finding on ChatGPT: "When ChatGPT search cites webpages, the pages it cites rank in traditional organic search positions 21+ for related queries almost 90% of the time." It added that Perplexity and Google's AI features also often cite lower-ranking pages, and that some AI systems cite positions 1–5 more often than positions 6–20. You can read the Semrush study in full.

Three details matter when quoting it:

• One subject area. Digital marketing and SEO topics, where content is plentiful and competition intense.
• Related queries. Rankings were checked for search terms related to the prompt, not the conversational prompt itself.
• One point in time. Mid-2025. ChatGPT's search has changed since then, and so may the pattern.

What do other studies find?

The studies are less contradictory than they look. Seer's Google matches had a median rank of 17 and an average of 28, which means many of them sat on Google's second or third page. A study that counts positions 21 and beyond as "low" would put a large share of those same citations in its low bucket. Checking rankings for a related keyword, as Semrush did, rather than the exact question pushes more pages into that bucket, because the keyword and the question retrieve different sets of pages.

All three agree that most ChatGPT citations are not Google's top results. Seer's comparison found a much closer match with Bing, 87% within Bing's top 20, which fits OpenAI's statement that ChatGPT search uses third-party search providers; but Ahrefs' larger July 2025 sample found ChatGPT's citations matched Bing's top 10 (8.1%) no more often than Google's (8.0%).

Why do ChatGPT's citations differ from Google's top results?

1. A different index. ChatGPT search draws mainly on third-party search providers and OpenAI's own crawler. Google data plays a smaller part: a 2025 report found OpenAI had used scraped Google results for some news and sports answers, and its shopping results closely track Google Shopping. For ordinary informational answers, Google's organic rankings are not the list ChatGPT picks from.
2. Different searches. ChatGPT turns a conversational prompt into its own search queries, which are often narrower or differently worded than the keyword a study checks.
3. Longer prompts. People write much longer prompts to ChatGPT than they type into Google. Specific pages that rank poorly for the short keyword can match the long question well.
4. Passage selection. ChatGPT cites the passages that answer parts of the question, which may sit on pages that rank lower overall.
5. Source mix. Answers often combine several sources, and some come from forums, documentation or smaller sites that do not compete for head terms.

How prompt length differs between ChatGPT and Google is covered in whether ChatGPT queries average 23 words compared with 4 for Google.

Does ChatGPT favor newer or niche content that does not rank on Google?

Partly. On freshness, an Ahrefs analysis of about 17 million citations found that AI assistants cite newer pages than Google's organic results do on average. ChatGPT cited the newest pages of the engines studied. So on average ChatGPT cites somewhat newer pages than Google ranks, but newness alone is not the lever: Ahrefs' 2026 analysis of 1.4 million ChatGPT prompts found the median cited page was about 500 days old.

On niche content, the effect comes from the questions rather than a preference. Specific, long questions are best answered by specific pages, which often rank poorly for broad keywords but well, in Bing or elsewhere, for the narrower searches ChatGPT runs. ChatGPT is not choosing pages because they do not rank on Google; it is choosing pages that answer the question, from a different index.

How should you read statistics like this?

Before quoting or acting on an AI citation statistic, check:

• Which engine and mode: ChatGPT with search, without search, or via the API.
• Which comparison: rankings for the exact prompt, a related keyword, or a topic.
• Which index: Google, Bing or another.
• Which subject area: results for SEO topics may not hold for local services or products.
• Sample and date: how many prompts and citations, and when.

A figure like "90% from position 21 or lower" can be true in its study and still misleading if applied to every business.

What does this mean for your strategy?

1. Check your Bing visibility. Since ChatGPT search draws on third-party search providers, Bing Webmaster Tools and IndexNow are worth covering for ChatGPT too.
2. Allow OpenAI's crawlers, OAI-SearchBot and ChatGPT-User, so your pages can be found and fetched.
3. Answer specific questions directly, since ChatGPT's searches are often narrower than your target keywords.
4. Keep pages current, given ChatGPT's tilt towards newer content.
5. Keep Google SEO going. It still drives Google's AI features and much of the wider web's visibility.

How ChatGPT, Perplexity and Google end up citing different sources for the same question is covered in why ChatGPT, Perplexity and Google cite different sources.

How can you check this for your own site?

Pick 20 questions your buyers ask. Run each in ChatGPT with search on, a few times, and record the cited URLs. For each cited URL, check its position in Google and in Bing, both for the full question and for the shorter keyword version. Note how many cited pages are in each engine's top 10 and top 20. Within an hour or two you will know whether the pattern in the studies holds in your market, and whether your pages lose out on Google rank, Bing rank or the answer itself.

What do people ask about ChatGPT citations and rankings?

Does ranking first on Google help in ChatGPT at all?

It helps indirectly, since strong pages tend to rank well everywhere, but Google rank alone predicts ChatGPT citations poorly.

Is ChatGPT search just Bing?

Not only. OpenAI says it uses third-party search providers and partner content, and its own crawler gathers pages. Studies disagree on how closely its citations match Bing's results.

Can a page with no Google rankings be cited by ChatGPT?

Yes, if it is indexed where ChatGPT searches and answers the question well.

Do these figures apply to Google's AI Overviews?

Not directly. AI Overviews draw on Google's own index, but even there the overlap with the top 10 has fallen: Ahrefs' 2026 update found about 38% of cited pages ranked in the top 10 for the query, down from 76% in 2025.